Zhe Xing, Han Yang, Yuan Lyu, Junqi Li, Xueli Tian, Qiao Shan, Wulong Liang, Weihua Hu, Shaolong Zhou, Xinjun Wang
PMLS independently predicted patient prognosis and demonstrated robust performance across all cohorts. It outperformed common clinical and molecular characteristics and most published signatures. Higher PMLS scores were associated with increased proliferation, inflammatory signaling, altered molecular features, and extensive immune-microenvironment remodeling. Simvastatin and fluvastatin were identified as potential therapeutic candidates for high-risk patients.
INTRODUCTION: Glioblastoma (GBM) is characterized by substantial inter- and intra-tumoral heterogeneity. A robust multigene model may help address this heterogeneity and improve prognostic stratification.
METHODS: We integrated 76 combinations derived from 10 machine-learning algorithms to develop a purificatory machine learning-derived gene signature (PMLS). The model was evaluated in 10 public cohorts comprising 1,097 patients and compared with common clinicopathological variables and 135 published prognostic signatures. Biological pathways, immune characteristics, and potential therapeutic agents associated with PMLS were further investigated.
RESULTS: PMLS independently predicted patient prognosis and demonstrated robust performance across all cohorts. It outperformed common clinical and molecular characteristics and most published signatures. Higher PMLS scores were associated with increased proliferation, inflammatory signaling, altered molecular features, and extensive immune-microenvironment remodeling. Simvastatin and fluvastatin were identified as potential therapeutic candidates for high-risk patients.
DISCUSSION: PMLS may provide a robust platform for prognostic stratification, biological interpretation, and the development of precision-treatment strategies for patients with GBM.